Vigensis
Book a Call
Menu
Sign up

VIA Research

AI-Native Product Development: Ergonomics Is the Real Advantage

People no longer operate every screen by hand. They delegate work and review it. That makes usability the advantage, for one task and for the whole life of a product.

· Andreas Müller, Product Maestro, Design Shaper, Tech Smith

  • AI-native
  • usability
  • product development

Most teams still treat AI as a feature. A chatbot in the corner. A “generate” button. A copilot bolted onto an existing workflow. That is useful. It is not AI-native product development.

An AI-native product starts from a different assumption. The person no longer operates every interface by hand. They work with a system that interprets intent, proposes actions, uses tools, remembers context, generates alternatives and completes work in several steps. The product is no longer a set of screens. It becomes a work environment that adapts.

That makes ergonomics and usability more important. ISO 9241-11 defines usability for systems, products and services in their context of use. For an AI-native product, that context now includes model uncertainty, automation, review, trust and human control.

AI-native does not mean AI everywhere

The worst AI products create extra work. They ask the person to prompt, inspect, correct, prompt again, verify, and then carry the result by hand into the system where the work happens. That is ergonomic debt.

A good AI-native product reduces the burden on the person in four dimensions: cognitive load, interaction friction, operational effort and organizational complexity. The interface understands intent, puts AI where the work happens, makes progress visible, and gives the person safe ways to approve, correct, undo or escalate.

The strongest AI-native products are easier to work with. That is the whole advantage.

From screens to intent loops

Traditional software is built around command execution: click, fill, submit, wait. AI-native software is built around intent loops: state a goal, let the system propose or execute a path, inspect the work, correct course, keep the context.

The pattern is already visible across the product development stack. Figma Make turns ideas and existing designs into working prototypes through conversation, so ideation and prototyping stay close together. GitHub Copilot’s cloud agent researches a repository, makes a plan, changes code on a branch and lets the developer review the diff before a pull request opens. Cursor runs agents across desktop, command line, web and mobile, beside manual coding in a familiar environment.

In VIA, the same loop runs at the altitude of a product. You bring an idea or a problem. VIA asks the questions it needs. A research specialist studies the problem, the market and the people who have the problem, and every claim shows its source. You read the decision on the opportunity and you go on or you stop. The run leaves a record: the goal, the sources, the plan, what was produced, where it was uncertain, and who decided.

The ergonomic shift is delegation with review. The AI does not do everything. The person stays in the loop at the points where judgment is theirs.

The usability problem: AI is probabilistic, but work is accountable

An AI-native product carries a basic tension. The model can reason, generate and act. The person remains accountable for the result. So the product must support confidence, correction and control.

A usable AI-native system answers five questions at all times:

  • What is the system doing?
  • Why is it doing that?
  • What information is it using?
  • What can I safely delegate?
  • How do I correct, undo or constrain it?

This is where many AI features fail. They give output without inspectability. They offer automation without recovery. They offer conversation without workflow state.

The better pattern is to design every AI action as a reviewable work unit: goal, sources, plan, proposed changes, risk areas and approval path.

From one task to the life of a product

A reviewable work unit answers the five questions for one task. A product lives for years. It is released, changed, audited, handed from one team to the next, and in a regulated place of work it must show at every step why it is the way it is.

The same unit answers for the lifetime, if the product keeps it. A product whose every change carries its goal, its sources, its plan, its tests and its approval is a product that can be reproduced and explained release after release. That is what audit trails, reproducibility and lifecycle discipline mean in practice. Regulated industries require them. Every product benefits from them.

This is the gap most AI tools leave open. They generate code for a task. They do not engineer, evolve and govern a product over its life. The hard part of AI-native product development is here, and so is the advantage.

Where VIA stands

VIA is an operating system for product development. It runs a product through its whole life, from the first question to a product in the market and on through every release after. It schedules the work through specialist agents and workflows. It keeps the record. It enforces the rules of the organization that runs it. It runs where that organization needs it, shared, dedicated or in its own custody.

It has an opinion about how a product should be built, and it enforces it: plan before you act, a record for every decision, tests with every release, a person at every point where judgment is theirs.

What comes out is yours. The code, the documents and the record of every decision belong to your organization.

The work of product development becomes something a product can carry

AI-native development blurs the line between product and service. Linear’s Customer Requests connects feedback from support, sales, CRM, email and Slack to product requests linked to issues and projects, so a team prioritizes from real demand. Discovery, once a service, runs inside the tool.

The same is true of the layers under an AI-native product. At the application layer, Vercel’s AI SDK gives a developer one toolkit for AI features and agents across React, Next.js, Vue, Svelte and Node.js. At the agent layer, OpenAI’s Responses API gives one interface for agent-like applications with built-in tools, multimodal input and tool calling. At the quality layer, LangSmith traces an LLM application from a single call to production-wide metrics.

Each layer takes a piece of work that used to be done by hand and makes it something the product carries. Product development as a whole is the next piece. Validating that a problem is real, defining the product, building it, releasing it and learning from its users can run inside one system, with the person deciding at each stage. That is the work VIA carries.

Design principles for ergonomic AI-native products

Put AI where intent appears. Do not send the person out of their workflow to “ask AI”. When they review customer feedback, the AI clusters, summarizes and connects it to the roadmap. When they edit a design, it understands the design context. When they review code, it works at the level of branches, diffs, tests and pull requests.

Make progress visible. The person sees the plan, the intermediate state, the assumptions, the sources and the proposed actions.

Design for interruption. Real work changes direction. An AI-native product needs partial completion, saved context, reversible actions and hand-over between people and agents.

Separate suggestion from execution. Some AI actions can be automatic. Some should be drafted. Some must wait for approval. The product sets the level by how reversible the action is and by what it can verify for itself.

Measure usability, and measure value. Track task completion, time to value, correction rate, review burden, escalation rate and the person’s confidence. Then take the last step. Measure what the tokens and the hours spent on a product return. An AI-native product that cannot show its value per token and per hour has not answered the question its buyer asks first.

The bottom line

AI-native product development does not replace product managers, designers, engineers or support teams. It redesigns the work environment so people operate at the level of intent while AI handles more of the translation, exploration, execution and synthesis.

The teams that win will not be the ones that add the most AI features. They will be the ones that make AI feel ergonomically natural: easy to start, easy to steer, easy to inspect, easy to correct, and safe to trust. For one task, and for the whole life of a product.

In the AI-native era, usability is the product.

Sources

A first version of this article appeared on Andreas Müller’s blog, Ergonomics & Usability, on 11 May 2026. This version is rewritten by Andreas with the VIA product trio to reflect where VIA stands today.

← All articles